Re: [sl4] Simple friendliness: plan B for AI

From: Matt Mahoney (matmahoney@yahoo.com)
Date: Sun Nov 14 2010 - 11:04:06 MST


Piaget Modeler wrote:
> Would Asimov's three laws be an easier starting point?
> If not, why not?

No, because the definition of friendliness has high algorithmic complexity.
There is no simple description. If there were, our legal system could be written
on one page.

Attempts to define it such as CEV ( http://intelligence.org/upload/CEV.html )
abstract out some of the complexity by not defining "human". The boundary around
the set of entities that deserve protection is already fuzzy. Otherwise there
would be no controversy over abortion, animal rights, or immigration policy. The
problem will get much worse when we have AI, uploads, and machine-human hybrids.

I estimate that the complexity of a definition of friendliness is on the same
order as the collective knowledge of the human race, about 10^17 to 10^18 bits.
But even if it were 10^9 bits, it would be beyond the capability of any person
to write it down. How do you define friendliness without a model of what
billions of people want and what constitutes a human mind?

 -- Matt Mahoney, matmahoney@yahoo.com



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:05 MDT